语音质量指标Python（PESQ、STOI、SegSNR、LLR、WSS、CD、LSD、CSIG、CBAK、COVL）_python计算pesq

在做语音增强的时候，论文中会提到一些语音质量的评价指标，无从下手，于是自己摸索了两天时间搞出了其计算的python版本。

1.安装对应的依赖库

#!/usr/bin/env python
# -*- coding: utf-8 -*-
'''
@Time    : 2020/9/13 16:53
@Author  : kingback
@Site    :
@File    : SEPM.py
@Software: PyCharm
'''


'''
the install order is :
因为一些函数包需要一些依赖，所以建议大家按照如下步骤安装操作


    0. All the package I use the commond "python setup.py install" to install to our machine.
    Maybe we also need to install other packages like :   cython ,numba   ,scipy  ,matplotlib ,numpy  ...
    0.说明，以下所有的安装包最好都再对应的github链接下载其源码，然后cmd进入到其对应的文件见，执行python setup.py install命令来进行安装

    1. install the gammatone from : https://github.com/detly/gammatone
    1. 第一步，先安装这个包，下载地址如上，下载好后进入文件夹执行命令python setup.py install

    2.install the SRMRpy from : https://github.com/jfsantos/SRMRpy
    2.这是需要安装的第二个包，因为这些包都是有依赖关系的，所以建议大家还是按照这个顺序来

    3.install the python-pesq from  :   https://github.com/vBaiCai/python-pesq
    3.这是第三个包

    4.install the pystoi from   :   https://github.com/mpariente/pystoi
    4.这是第四个

    5.install the pyseqm from   :   https://github.com/schmiph2/pysepm
    5.这是第五个

'''
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36

如果大家链接github速度较慢，可以访问那五个安装包，提取码：r809

2.环境搭建

环境搭建的话，博主是使用conda新创了个python3.6的虚拟环境，然后再搭建的这个语音质量指标环境，以下是搭建过程及步骤：

1.创建新环境 conda create -n tp python=3.6
2.激活环境 activate tp
3.安装必备包（因为下边要安装的包需要这些依赖，如果使用pip源的话就很慢，所以在这里提前下载下来,这三个包必须提前安装哈）
pip install matplotlib -i https://pypi.tuna.tsinghua.edu.cn/simple
pip install scipy -i https://pypi.tuna.tsinghua.edu.cn/simple
pip install numba -i https://pypi.tuna.tsinghua.edu.cn/simple
4. 进入到gammatone_master里边执行python setup.py install命令
5. 安装成功，如下图所示
1
2
3
4
5
6
7
8

6. 紧接着同样的步骤安装SRMRpy,记得切换到tp环境哦

然后执行python setup.py install进行安装，出现以下界面就是安装成功了

7.接下来安装pesq,和上边的步骤一样，在对应的文件夹下进入cmd命令行，然后切换到tp虚拟环境，执行命令python setup.py install

8.和上边的操作一样，安装第四个包pystoi

9.步骤同上，安装第五个包pysepm

3.代码部分，计算语音的质量指标

因为新环境并没有soundfile包（用于读取语音文件）和tqdm包（用于显示进度），所以想要以下代码，还需要安装这俩包，执行以下命令

pip install soundfile tqdm
1

以下就是安装成功的结果啦

都安装完以后，就可以测试语音的质量指标啦，以下就是全部的代码：

import pysepm
import os
import numpy as np
import soundfile as sf
from tqdm import tqdm

clean_wavs = 'clean_testset_wav_16k/'#这个路径，大家根据自己的需求进行修改
denoised_wavs = 'cleaned_testset_wav_16k_48700/'#同理进行修改


def file_name(file_dir):
    L = []
    for root, dirs, files in os.walk(file_dir):
        for file in files:
            if os.path.splitext(file)[1] == '.wav':
                L.append(os.path.join(root, file))
    return L


# get wav_lists
clean_lists = file_name(clean_wavs)
denoised_lists = file_name(denoised_wavs)
# Package files
zipped = zip(clean_lists, denoised_lists)

scores = []
CSIGs = []
CBAKs = []
COVLs = []

for (clean_wav, denoised_wav) in tqdm(zipped, 'the progressing ...'):
    # Gain speech parameters
    ref, sr0 = sf.read(clean_wav)
    deg, sr1 = sf.read(denoised_wav)



    '''
    # Method 1: SNRseg (分段信噪比)
        # from pysepm Call SNRseg to calculate its metrics
        # in this case we can choose our frame length =0.03*1000=30 ms , and the overlap =30 ms *0.75 =22.5 ms
        # 得分越高，质量越好
    '''
    # score = pysepm.SNRseg(ref, deg, sr0)



    '''
    # Method 2: llr (对数似然比测度)
        # 
        # 得分越高，质量越好
    '''
    # score = pysepm.llr(ref, deg, sr0)



    '''
    # Method 3: WSS (加权谱倾斜测度)
        # 
        # 得分越低，质量越好
    '''
    score = pysepm.wss(ref, deg, sr0)


    '''
    # Method 4: STOI (可短时客观可懂)
        # 
        # the score from 0-1 . The higher the score, the better the performance.
        #得分范围0~1,得分越高，质量越好
    '''
    # score = pysepm.stoi(ref, deg, sr0)


    '''
    # Method 5: PESQ
        # when I try this commond , I faced some troubles   , so finally I gave up this commond,
        # use the PESQ.py to instead
        得分范围从 -0.5~ 4.5，得分越高，效果越好
    '''
    NaN,score = pysepm.pesq(ref,deg,sr0)



    '''
    # Method 6: CD (Cepstrum Distance)
        # 
        # 数值越高，得分越高.
    '''
    # score = pysepm.cepstrum_distance(ref, deg, sr0)


    '''
       # Method 7: LSD (对数谱距离)
            # This method I use the LSD.py to calculate the distance 
            # 数值越小，得分越高      
    '''


    '''
    Method 1 - 7 use this score to print
    '''
    # score append to scores
    scores.append(score)


    '''
    # Method 8: Composite
        # In this method , It comes some errors, if you want to solve the error ,  see the step 8 in this file.
        # CSIG , CBAK , COVL all from 1 - 5 , The higher the score, the better the performance.
    '''
    # CSIG,CBAK,COVL = pysepm.composite(ref, deg, sr0)
    # CSIGs.append(CSIG)
    # CBAKs.append(CBAK)
    # COVLs.append(COVL)


# print(scores)
print('The average SegSNR evaluation is :   ', sum(scores) / len(scores))
# calculate the standard deviation & variance of the scores
print('The standard deviation is :   ', np.std(scores))
print('The variance is :   ', np.var(scores))


# print(CSIGs,CBAKs,COVLs)
# print('The average CSIG evaluation is :   ', sum(CSIGs) / len(CSIGs))
# print('The average CBAK evaluation is :   ', sum(CBAKs) / len(CBAKs))
# print('The average COVL evaluation is :   ', sum(COVLs) / len(COVLs))
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127

4.可能出现的问题及解决措施（应该不会出现）

4.1 错误 ModuleNotFoundError: No module named ‘pesq’

 6. if  you have finished the last step  ,don't forget to change the next step's setup.py .
    This is done so that you do not have to repeatedly download the installed files.
    And there's also the possibility of an error.
   如果你在程序运行过程中出现以下问题，可以参考以下解决方案。

7.  如果你的代码在执行过程中遇到以下问题，可以参照此教程，
       "
        File "/home/king/anaconda3/envs/tf_gpu/lib/python3.6/site-packages/pysepm-0.1-py3.6.egg/pysepm/__init__.py", line 4, in <module>
        File "/home/king/anaconda3/envs/tf_gpu/lib/python3.6/site-packages/pysepm-0.1-py3.6.egg/pysepm/qualityMeasures.py", line 4, in <module>
        ModuleNotFoundError: No module named 'pesq'
        "
		找到你的环境中的这个pysepm/qualityMeasures.py文件，当然路径肯定和我这不一样哈
        you can go to your file path "/home/king/anaconda3/envs/tf_gpu/lib/python3.6/site-packages/pysepm-0.1-py3.6.egg/pysepm/qualityMeasures.py"

		在源码中找到如下命令:
            #import pesq as pypesq # https://github.com/ludlows/python-pesq
        替换为 :
            import pypesq
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18

4.2 测量语音的PESQ指标和CBark，CSIG，COVL时可能遇到的问题及解决措施

    9.  if you run the method pesq & Composite , it may be come some errors because we haven't introduced PESQ packages properly yet.
    10.如果你在测量语音的pesq指标，和Composite 指标（即CBark,CSIG，COVL） ,如果出现问题，建议使用如下解决方案
    	打开以下文件
        you can go to your file path "/home/king/anaconda3/envs/tf_gpu/lib/python3.6/site-packages/pysepm-0.1-py3.6.egg/pysepm/qualityMeasures.py"
       （1） 操作1 #在文件开头添加以下命令
            "from pypesq import pesq as pq"
            
        (2).操作2 ,替换347-359行的pesq函数
            "
                def pesq(clean_speech, processed_speech, fs):
                    if fs == 8000:
                        mos_lqo = pq(clean_speech,processed_speech, fs)
                        pesq_mos = 46607/14945 - (2000*np.log(1/(mos_lqo/4 - 999/4000) - 1))/2989#0.999 + ( 4.999-0.999 ) / ( 1+np.exp(-1.4945*pesq_mos+4.6607) )
                    elif fs == 16000:
                         mos_lqo = pq(clean_speech,processed_speech, fs)
                        pesq_mos = np.NaN
                    else:
                        raise ValueError('fs must be either 8 kHz or 16 kHz')
                    return pesq_mos,mos_lqo
            "
        (3).保存并且退出
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21

5.一劳永逸的方法解决各种错误（简单粗暴）

博主在近期重新搞了搞这个，其实如果重新搭建环境并且按照 第二部分和第三部分环境搭建的话是没有出现问题的，且经过博主本人再三检测，以上所有音频指标均可正常检测，如果大家在使用过程中出现错误，为方便大家，我将该conda虚拟环境传输到百度云网盘，以提供给大家使用，下载完成后，将tp环境放置到conda安装路径下的envs文件夹下，之后选择虚拟环境，就可以直接在这个虚拟环境上进行操作啦。百度云链接，提取码：ggek

6.测试数据集

本博客中用于检测的音频数据，我也一并上传至百度网盘，大家也可拿来进行测试数据集链接，提取码：ybjp

7.参考资料

1. https://github.com/schmiph2/pysepm/blob/master/examples/examplesForCalculatingMeasures.ipynb (How to use the packages of SEPM)
2. https://zhuanlan.zhihu.com/p/190146707 (音频质量客观评价指标)

8.修改说明